walkthrough · rationale · spec

List data structures for weighted randomisation, implemented (eventually) in every programming language I’ve learnt.

let descriptors = wlist![
    (10, "cool".to_owned()),
    (5,  "awesome".to_owned()),
    (2,  "elegant".to_owned()),
    (1,  "beautiful".to_owned()),
];

let words = descriptors.select_random_values_unique()
    .rng(&mut rand::rng())
    .count(2)
    .call();

println!("Rust is {} and {}", words[0], words[1]);
// => Rust is awesome and elegant

greetings = WeightedList((20, "sup"), (2, "salutations"))

print(greetings.select())
# => sup

WeightedList<string, int> greetings = new((20, "sup"), (2, "salutations"));

Console.WriteLine(greetings.GetRandomValue());
// => salutations

let greetings = new FrozenWeightedList([20, "sup"], [2, "salutations"]);

console.log(greetings.sample_value());
// => sup

greetings :: WeightedList String Int
greetings = newWeightedList [(20, "sup"), (2, "salutations")]

main :: IO ()
main = print (randomValue greetings)
-- => salutations

(working on it!)

• Weighted randomised selection with constraints such as no-replacement, multi-take, unique-only
• Ergonomic interface targeted for each language
• Utility methods for manipulating values and weights
• In-place and out-of-place variants of methods
• Conversions to and from other data types

• Slice indexing¹

I made this class for weighted randomisation, where each element in a collection has a different chance of being selected – the greater an item’s weight, the higher the chance it is selected. This is super common in games for reward systems, displaying messages, etc.

crates.io · documentation

> cargo add weighted-list

use weighted_list::{ WeightedList, FrozenWeightedList };

npm

npm install @sup2.0/weighted-list

import { FrozenWeightedList } from "@sup2.0/weighted-list";

All you need is the weightedlist.py file, which contains the WeightedList class with all the functionality. Simply import it, and you’re ready to go!

from weightedlist import WeightedList

See walkthrough for a tutorial, or examples for examples.

All the code is contained within the WeightedList.cs file. You might also need the weighted-list.csproj file. If you want the entire solution, you can download the repo and extract the c-sharp/ folder.

For a tutorial, see walkthrough.

import WeightedList qualified as WL
import WeightedList (WeightedList)

A WeightedList stores ‘weighted items’ instead of raw values. Each item has a weight and value attribute.

WeightedList using weighted indexing. To illustrate with an example:

wl = WeightedList((1, "qi"), (2, "sup"), (5, "shard"))

assert wl[0].value == "qi"
assert wl[1].value == "sup"
assert wl[2].value == "sup"
assert wl[3].value == "shard"
assert wl[4].value == "shard"
assert wl[5].value == "shard"
assert wl[6].value == "shard"
assert wl[7].value == "shard"

wl[8]  # IndexError

In essence, you can think of each item as being repeated a number of times equal to its weight. The ‘length’ of the list is the sum of the weights.

Hence, selecting a random value using weighted randomisation is as simple as picking a random index between $0$ and the length of the list!

Back when I was picking up the ropes of Python, I was working on a project which featured randomisation, and, like any game developer, I thought it’d be cool to give each outcomes different probabilities of occurring. At first, I achieved this behaviour by duplicating items, but I quickly realised the numerous issues with this.

And so, I set out to write my own class, which I’d never really needed to do up until that point. I thought it’d be a great exercise in learning Python – and it very much was, teaching me tons about object-oriented programming, dunder methods, generators, etc. It was also my first experience of conscientiously writing code that wasn’t exclusively for myself, which helped me understand the importance of consistency and clarity, and above all, documentation.

A couple years later, I’ve come back to do the same in C#, this time also adding several features I always intended to add but never did – especially non-integer weights, which allows the class to truly embrace its usage as representing probabilities. Trying to translate Python into C# was an interesting experience,² and helped highlight some important differences between the languages that I would otherwise not have found out.

A few more years later, I’m back to do the same in Haskell and Rust (and also finish off the TypeScript and Ruby implementations that I started but never finished). Damn I love this project. Seriously, it never fails to raise so many questions about a language’s mechanics and quirks that I would never encounter otherwise.

I mean yeah, a whole several-hundred-lines class to handle one thing is probably overkill... it’s more an exercise and proof-of-concept.

Regardless, I’ve used my own code³ in at least 2 major projects (PENGUIN^↗ and Algorhythm^↗), so I can definitely say it’s been useful to me!

1. documentation
2. line breaks
3. utility

Particularly documentation. That stuff just eats the line count. Also, implementing something as complex as an enumerable container requires a lot of methods, operators, interfaces and delegation. And in C# you've even got overloading to account for as well.

In all honesty, I don’t know. I’m slowly adding benchmarks to test different approaches.

I have my own particular preferences when it comes to coding in Python, which I explain fully here^↗.

I’m a C# programmer, what can I say :P

I’m a Python programmer, what can I say :P

I used snake_case before, and ngl, in Haskell you kinda need the camelCase to keep things readable without parentheses...

Any feedback, suggestions or improvements are definitely welcome!